Mini Project III:

Patterns in Banking Behavior

Ali Bahrami & Chunshan Feng

November 16, 2021

Goal

Determine which client segments are bringing in the most revenues and risks for the bank

• How do banks make money?
  • Net Interest Income: on interests they earn by lending borrowed money to clients
  • Interchange income: through fees paid by merchants per transaction
• How do banks lose money?
  • Default on loans (i.e. credit card loans, mortgage loans, business loans, etc.)

Agenda

1. Do an expolartory analysis of the bank's customer base using demographics data:
   • Income
   • Gender
   • Number of children
2. Customer segmentation using income in the following segments:
   • revenue: total interests paid, credit balance owed
   • revenue: total transactions made
   • risk: percent of credit spent
3. PCA Analysis and Radar Charts

Explore the Dataset

# Age Distribution
fig = px.histogram(df_customer.age, x="age", title='Age Distribution', nbins=16)
fig.show()

# Gender Distribution
fig = px.histogram(df_customer.gender, x="gender", title='Gender Distribution', width = 600)
fig.show()

# Number of children
fig = px.histogram(df_customer.nbr_children, x="nbr_children", title='Number of Children per Customer', width = 600)
fig.show()

# Income disparity by Gender
fig.show()

Income Distribution by Age and Income

Clusters by Income and Transaction Amount

Group 3 (turquoise): Rewards products

Clusters using Credit Balance

Group 3 (turquoise): Premium Cashback and Rewards Card with first-year fee rebate

Customer Segments - Net Worth

Net Worth = (Checking + Savings) - Credit

Customer Segments - Risk

Group 1 (turquoise): Credit limit increase offer

PCA Analysis

# Multi-dimensional Analysis
df_segmentation

	gender	income	credit_interests	total_trans	net_worth
0	1	50890	361.26	84	1254.83
1	0	10053	14.81	54	1847.77
2	0	22690	56.93	94	-1054.51
3	1	6605	13.63	15	-134.13
4	0	55888	248.77	124	479.08
...	...	...	...	...	...
452	1	9271	79.61	62	-596.25
453	0	10244	12.21	130	-175.02
454	0	19863	41.27	18	-842.24
455	0	39942	0.00	17	0.00
456	1	142274	963.28	135	-5646.76

457 rows × 5 columns

Radar Chart

Explained Variance

pd.DataFrame(pca.explained_variance_ratio_).plot.bar()
plt.legend('')
plt.xlabel('Principal Components')
plt.ylabel('Explained Varience');

Post-PCA K-Means Clustering

Post-PCA Cluster